Incorporating Adaptive Sparse Graph Convolutional Neural Networks for Segmentation of Organs at Risk in Radiotherapy

Table 1

Comparison between the vanilla 3D U-Net and the network applied with the proposed ASGCN.

ROI	DSC of 3D U-Net	DSC of ASGCN	HD of 3D U-Net	HD of ASGCN
Constrictor naris	0.7565	0.7722	2.1913	1.7650
Eyes	0.8791	0.8951	2.0778	1.1163
Lens	0.6160	0.6495	4.6414	2.4956
Optic nerves	0.6077	0.6418	2.5521	2.3080
Temporal lobes	0.8595	0.8690	3.0174	2.1383
Thyroids	0.7826	0.8089	2.9000	2.0280

The upward and downward arrows indicate that the higher and lower are better, respectively. The bold value denotes better performance.